DeepMind published predicted effects for about nine billion single-letter DNA changes

Google DeepMind unveiled AlphaGenome Atlas on Tuesday, a searchable catalog that predicts the molecular effects of every possible single-letter DNA change in the human genome, Robert Hart at The Verge writes. Company language calls it a predictive map of every DNA letter change.

DNA is written in four chemical letters, usually shortened to A, C, G, and T. The human genome holds about three billion letter pairs. Flip one letter and you get roughly nine billion possible single-letter substitutions, point mutations that can be harmless, ordinary human difference, or part of disease.

Atlas predicts what each of those flips does at the molecular level, for example how much of a protein is produced. Researchers call it the most comprehensive catalogue of how genetic mutations affect molecular biology. That superlative is the company's claim.

Access routes include a web portal, a skill in Antigravity, Google's product for AI agents, and the AlphaGenome interface. A Variant Impact Score, AVI, ranks variants using other Google models so researchers can sort billions of possibilities.

The project builds on AlphaGenome, a model DeepMind released last year, and AlphaMissense, an earlier tool focused on protein-altering small mutations. Atlas extends predictions across the genome, including noncoding stretches that do not make proteins directly but can turn genes up or down.

Ziga Avsec, DeepMind's genomics lead, told a press briefing that the underlying model was already out. Precomputing and analyzing this many variants took time because the space is so big.

Training used public human and mouse genome databases. The resulting dataset is about one petabyte, roughly a million gigabytes.

Atlas is available to researchers for noncommercial use via the website starting today. Commercial use on Google Cloud is coming soon.

Disease-treatment language in the coverage is forward-looking framing. Atlas does not cure anything on its own.

Nine billion flips is too large to walk by hand. The map is the product. Whether labs can turn ranked molecular predictions into therapies is the next, slower work.

Sources

• The Verge — Google / DeepMind AlphaGenome Atlas